A empirical mode decomposition approach to feature extraction of ship-radiated noise

How to obtain effective, reliable characteristic parameter from the limited measured data is a question of great importance in feature extraction. Based on self-adaptive filter action of empirical mode decomposition (EMD) method, this paper drew statistic centre frequency of spectrum of intrinsic modes as new line spectrum characteristic of underwater acoustic signal, and adopted the law of nearest neighborhood to recognize. The characteristics used in this method include the high-frequency characteristics which other spectral analysis methods neglect, so it can get higher discrimination when ten types of objects are classed despite small sample volume and less data amount.